Fig. 2
From: MethylNet: an automated and modular deep learning approach for DNA methylation analysis
Age Results on Test Set (n = 144): a Age predictions derived using the Horvath, Hannum, and MethylNet estimators are compared to the true age of the individual, the predicted ages are plotted on the x-axis, the actual ages on the y-axis, and a line was fit to the data for each estimator; b Comparison of MethylNet Age estimates on Test Set (n = 144) to Horvath and Hannum Age Estimators. 95% confidence intervals for each score were calculated using a one thousand sample non-parametric bootstrap; c Bar chart depicting the overlap of CpGs important to MethylNet and Hannum age estimators where one thousand CpGs with the highest SHAP scores per 10-year age group are divided by the total number of Hannum CpGs that passed QC; d Hierarchical clustering using the correlation distance between SHAP CpG scores for age groups across all CpGs. The linkage is found between similar age groups